Dynamic runoff connectivity of overland flow on steep forested hillslopes: Scale effects and runoff transfer

Authors

  • Takashi Gomi,

    1. Department of International Environmental and Agriculture Science, Tokyo University of Agriculture and Technology, Tokyo, Japan
    Search for more papers by this author
  • Roy C. Sidle,

    1. Geohazards Division, Disaster Prevention Research Institute, Kyoto University, Kyoto, Japan
    Search for more papers by this author
  • Shusuke Miyata,

    1. Laboratory of Erosion Control, Division of Forest and Biomaterial Science, Graduate School of Agriculture, Kyoto University, Kyoto, Japan
    2. Now at Department of International Environmental and Agriculture Science, Tokyo University of Agriculture and Technology, Tokyo, Japan.
    Search for more papers by this author
  • Ken'ichirou Kosugi,

    1. Laboratory of Erosion Control, Division of Forest and Biomaterial Science, Graduate School of Agriculture, Kyoto University, Kyoto, Japan
    Search for more papers by this author
  • Yuichi Onda

    1. Department of Integrative Environmental Sciences, Graduate School of Life and Environmental Sciences, University of Tsukuba, Tsukuba, Japan
    Search for more papers by this author

Abstract

[1] Both scaling effect and connectivity of overland flow were examined in steep hillslopes covered by (1) Japanese cypress (hinoki, Chamecyparis obtusa) plantations with sparse understory vegetation, (2) hinoki plantations with fern understory vegetation, and (3) deciduous forests. Two sizes of plots were installed for monitoring overland flow: small (0.5 × 2 m) and large hillslope scale (8 × 24–27 m). For all hillslopes, measurable amounts of overland flow occurred during storms. Runoff coefficients of large plots (0.1–3%) were consistently smaller than those of small plots (20–40%). Estimated runoff flow lengths at the hillslope scale were based on runoff coefficients from small plots and were used to calculate runoff volume from large plots. Then we compared the differences between observed and estimated runoff volumes of large plots. Estimated runoff from large plots was smaller than observed runoff in hinoki slopes with sparse understory vegetation. Greater amounts of observed compared to estimated overland flow suggest that more runoff occurred from hillslopes with sparse understory. In contrast, estimated overland flow was larger than observed runoff from the deciduous forest, implying greater opportunities for infiltration compared to hinoki hillslopes. Comparison of estimated versus observed overland flow for successive 5 min intervals during storms indicates that runoff networks expand upslope during short and intense precipitation periods. Our examination and comparison of storm runoff from small and large plots facilitate better understanding of runoff mechanisms, scaling effects in hillslopes, and connectivity of the overland flow network.

1. Introduction

[2] Runoff generation on hillslopes is highly variable in both spatial and temporal scales [Wood et al., 1986; Abrahams et al., 1989; Sidle et al., 2000; Godsey et al., 2004]. Plot-scale studies have measured overland flow and infiltration capacity for various soil surface conditions, including agricultural land, grassland, and road surfaces [Duley and Ackerman, 1934; Chaplot and Le Bissonnais, 2000; Ziegler et al., 2000]. When comparing plots of different lengths, unit area volumes of overland flow tend to decrease with increasing plot length [Poesen, 1992; Wainwright and Parsons, 2002; Parsons et al., 2006]. Joel et al. [2002] found that large plots produced only 40% of the runoff measured from small plots. In laboratory experiments, overland flow from plots decreased exponentially with increasing plot length [Bryan and Poesen, 1989; Stomph et al., 2002]. Doerr et al. [2003] noted that hydrophobic forest topsoil exerted the greatest influence on overland flow at smaller scales with these effects decreasing with increasing area. In desert shrub plots ranging from 2 to 27.8 m in length, a significant inverse relationship between runoff coefficient and plot length was observed [Parsons et al., 2006]. Scaling effects are more evident when comparing runoff coefficients between plots and catchments [Cammeraat, 2002; Cerdan et al., 2004]. Despite these findings, such scaling effects were not evident in some other studies or even increases in overland flow were found with increasing slope length [Fox et al., 1997; Chaplot and Le Bissonnais, 2000; Joel et al., 2002]. Nevertheless, scaling effects suggest that the occurrence and transfer of overland flow is spatially variable and the order of magnitude of infiltration and runoff in small plots is not always the same as in large plots and catchments [Sivapalan and Wood, 1986; Doerr et al., 1998; Parsons et al., 2004].

[3] Scaling effects of hillslope overland flow potentially change with time because of changes in rainfall intensity and soil surface conditions [van de Giesen et al., 2000; Stomph et al., 2002; Wainwright and Parsons, 2002; Vigiak et al., 2006]. Spatial variability of connectivity of runoff on hillslopes and the subsequent development of flow networks affect scaling [Yair, 1992; van de Giesen et al., 2000; Doerr et al., 2003; Stomph et al., 2002]. Because the occurrence of overland flow is directly related to rainfall intensity and infiltration capacity [Horton, 1933], the development of overland flow and runoff transfer may differ with changes in rainfall intensity and soil moisture [Dunne et al., 1991]. Overland flow transfer through a connected network also depends on soil surface conditions, such as availability of organic matter, surface soil sealing, stone cover, variation of surface hydraulic condition, and soil hydrophobic conditions [Moore and Larson, 1979; Sharma et al., 1980; Doerr et al., 2003; Vigiak et al., 2006]. Larger stones distributed on the soil surface facilitate discontinuous overland flow transfer [Lavee and Poesen, 1991; Yair, 1992]. Connectivity of overland flow networks may exhibit threshold behavior because initiation of overland flow requires a certain level of rainfall infiltration, filling of soil depressions, and water depth relative to soil surface roughness. Vegetation cover potentially affects overland flow generation and transfer because infiltration and slope roughness vary among vegetation types [El-Hassanin et al., 1993; Mueller et al., 2007]. Yet, few studies have demonstrated and quantified the connectivity of overland flow and the dynamic changes in overland flow for different cover conditions on hillslopes during storms.

[4] Occurrence of overland flow and soil surface erosion in monoculture stands of Japanese cypress (hinoki, Chamecyparis obtusa) forests have been noted [Miyata et al., 2007; Sidle et al., 2007; Nanko et al., 2008]. Greater volumes of overland flow occurred on hillslopes with sparse understory vegetation than hillslopes covered by ferns and litter layers during most storms [Sidle et al., 2007]. Soil water repellency is partly attributed to overland flow production on Japanese cypress hillslopes [Miyata et al., 2007; Kobayashi and Shimizu, 2007]. Under such soil surface conditions, splash erosion and soil surface sealing or crusting may occur [Onda and Yukawa, 1994; Nanko et al., 2008], thereby reducing infiltration capacity and generating overland flow.

[5] Understanding spatial and temporal variations in overland flow and its transfer are essential for examining its contribution to catchment storm runoff. Scaling issues are major concerns related to the continuous and discontinuous nature of overland flow and pathways. Hydrologic and geomorphic linkages from hillslopes to channels are important factors for understanding material dynamics in headwater catchments [Sidle et al., 2000; Gomi et al., 2002]. Better conceptualization of processes including flow variability and pathways is essential for improving catchment runoff modeling [Grayson and Bloschl, 2000]. Therefore, we monitored overland flow generation in different sizes of hillslope plots with various types of understory and overstory vegetation cover. By comparing runoff volumes and response, we examined (1) scaling effect of overland flow in forested hillslopes, (2) effect of vegetation condition on the scaling effect of overland flow, and (3) the connectivity of overland flow transfer among different vegetation cover conditions and during different storm conditions. Our experimental approach at the large and small hillslope scales facilitates the understanding of runoff connectivity and catchment-scale linkages between hillslopes and channels.

2. Concept of Hillslope Overland Flow and Its Transfer

[6] Since Horton proposed the concept of infiltration excess flow, various conceptualizations related to the occurrence and transfer of overland flow had been demonstrated [Horton, 1933; Emmett, 1978]. Delayed overland flow runoff was identified on the basis of changes in soil structure, detention storage, and roughness of the soil surface. Such delayed response was modeled using a kinematic wave approximation [Emmett, 1978]. Increasing the flow depth alters nonerosive laminar flow to more turbulent flow which induces rill and gully erosion. Overland flow is generally expressed as the length of the horizontal projection of an overland flow plane (L), flow depth (h), mean flow velocity (v), mean slope gradient (s), excess rainfall intensity (i), and discharge per unit width (w) [e.g., Julien and Moglen, 1990]. Excess rainfall intensity is expressed as rainfall intensity minus infiltration capacity. Thus, conservation of mass for one dimensional overland flow can be expressed as

equation image

where Q is defined as Q = αhβ, α is a function of Manning's coefficient, and β is an empirical coefficient. Equation (1) can address variable widths of overland flow for steady state rainfall and infiltration [Julien and Moglen, 1990].

[7] Despite this theoretical development and conceptualization, many variables can contribute to heterogeneous patterns of overland flow and its transfer across hillslopes [Emmett, 1978; Abrahams et al., 1989]. Such overland flow is typically not steady state because rainfall intensity varies spatially and temporally during the storm [e.g., Wainwright and Parsons, 2002]. Under vegetation cover, spatial and temporal variation of precipitation tends to be greater and skewed [Wilson et al., 1979; Konishi et al., 2006]. More realistic approximations for such dynamic rainfall conditions include variable width (w) and flow depth (h) in equation (1), depending on the amount of excess rainfall intensity (i), and changes in the length of overland flow (L). Thus, change in the length of the overland flow plane appears to be important for modeling runoff on natural hillslopes using a kinematic wave approximation.

[8] Routing of overland flow and transport distance are interrelated phenomena. To conceptualize overland flow and its transfer more realistically for specific hillslopes, runoff occurrence from isolated areas and the subsequent development of overland flow networks should be considered (Figure 1). Topographic irregularities are sufficient to concentrate runoff [Emmett, 1978; Dunne et al., 1991; Darboux et al., 2001]. Spatial variation of infiltration capacity and saturated hydraulic conductivity is associated with spatial variability of soil moisture, vertical preferential flow (e.g., vegetation roots), and soil hydrophobic conditions, as well as natural variability in soil physical properties [Sivapalan and Wood, 1986; Sidle et al., 2000; Weiler and Neaf, 2003]. Thus, overland flow generated on hillslopes is discontinuous when local infiltration and ponding dominate relative to excess rainfall (Figure 1a). Changes in input and/or transfer factors (e.g., precipitation intensity and/or soil sealing) may promote continuous flow on the soil surface once sufficient preferential flow paths develop because of spatial variability of flow depth and velocity (Figure 1b). When the discontinuous overland flow dominates on hillslopes (Figure 1a), the mean transfer distance and amount of overland flow captured at the bottom of a hillslope segment is relatively independent of slope length. For this condition, the runoff coefficient decreases with increases in slope length. In contrast, for continuous overland flow (Figure 1b), both the transfer distance and the amount of overland flow depend on slope length.

Figure 1.

Concept of overland flow connectivity at the hillslope scale and method for estimating runoff transfer distance. Overland flow generated on natural hillslopes typically occurs on specific flow paths. Depending on rainfall intensity and infiltration capacity, overland flow on given hillslopes is discontinuous. Such discontinuous portions potentially connect when the rainfall-infiltration ratio changes (e.g., increases in precipitation intensity).

[9] For our site, we hypothesize that overland flow generation on forested hillslopes is predominantly discontinuous on the patches of hillslope (Figure 1a) because root networks of trees and understory vegetation facilitate infiltration and there was no evidence of concentrated surface flow paths (e.g., rills) on hillslopes [Sidle et al., 2007]. Scaling effects whereby runoff coefficients decrease with increasing slope length can occur because of overland flow discontinuities. To test this hypothesis, we estimated the mean transfer distance of overland flow for different sizes of plots. The mean contributing area of overland flow (acr) can be estimated on the basis of runoff coefficients during specified periods of a storm (e.g., 5 min or the total event). For a specific plot size with a runoff coefficient Rc, rainfall and runoff from the lower portion of plot A (in this case, the small plot) occurs as overland flow (Figure 1c). We employed the simplifying assumption that measured runoff occurred from the lower half of the plot. More realistically, runoff may have arisen on a few patches within the plot (Figure 1c) and subsequently delivered to the outlet without infiltration into permeable areas [Emmett, 1978]. Thus, our assumption of mean contributing area was applied from the lower end to the plots, although it is recognized that heterogeneous flow networks develop within the plots. Therefore, the contributing area acr of plot A can be expressed as

equation image

where as is horizontal projected area of the lower portion of plot A. Thus, the mean transfer distance (projected mean length) of overland flow (lcs) can be derived as

equation image

where w1 is the width of plot A.

[10] On the basis of our hypothesis, the mean transfer distance (lcs) of overland flow can be applied to the other sizes of plots (e.g., plot B, large plot in Figure 1c). If plots A and B have similar topography and surface vegetation, the mean runoff contributing area of plot B is

equation image

where w2 is the width of the plot B. Therefore, the estimated runoff coefficient for plot B (R*) is calculated as

equation image

where a2 is the projected horizontal area of plot B. The estimated runoff amount (Q*, mm) is

equation image

where p is the precipitation for a specific period.

[11] By comparing estimated Q* and observed Q from plot B (Figure 1c), the following three cases can be tested: (1) Q* = Q, our hypothesis for mean transfer distance is valid and acceptable for all plots; (2) Q* > Q, more infiltration occurs along the hillslope and the mean transfer distance is overestimated; and (3) Q* < Q, more runoff occurs than predicted by small plots, and the mean transfer distance is underestimated. Thus, comparison between observed and estimated runoff based on our runoff distance assumption can be used as an index of the connectivity of overland flow rather than actual flow distance on given hillslopes and specific storms. For instance, greater observed overland flow than estimated overland flow indicates the occurrence of connected overland flow. On the basis of these three scenarios, relative continuity or discontinuity of overland flow can be examined.

3. Study Site and Field Methods

[12] This study was conducted within a steep 4.9 ha forested catchment in central Mie Prefecture (latitude, 34°21′N; longitude, 136°25′E; altitude ranged from 100 to 260 m), south central Japan (Figure 2). Climate in this area is moist and temperate with a mean annual precipitation of about 2000 mm and mean annual air temperature of 14°C. Two storm periods typically dominate: the Baiu season from late May through June, and the typhoon season from late August through October, although actual precipitation amounts during these storm periods vary widely from year to year. Soils in the catchment are brown forest soils ranging in depth from 0.6 to 1.8 m. Soils are relatively shallow at lower hillslope positions and thicker at midslope and near the ridge line. The combined A and B horizons are about 25 to 30 cm thick, underlain by a C horizon typically >35 cm thick. The catchment is deeply incised with a dominant hillslope gradient of 35°. The forest is a 40-year-old stand of Japanese cypress (hinoki, Chamecyparis obtusa) with a few small inclusions of Japanese cedar (sugi, Cryptomeria japonica). Dominant understory vegetation is fern (Gleichenia japonica) and evergreen shrubs (e.g., Cleyera japonica). A small area just outside of this catchment with deciduous forest cover was selected as a comparison site.

Figure 2.

Location and topography of study catchments and plots.

[13] Three types of planar hillslope segments with different stand densities and understory vegetation cover were selected (Table 1). Hillslope segments with little topographic roughness (e.g., extensive woody debris, boulders, slope breaks) were selected for plot installations (Figure 3). Hillslope 1 was covered by dense (4500 stems ha−1) cypress with sparse fern understory vegetation. Hillslope 2 represented a cypress stand with 1500 stems ha−1; thus the average stand diameter was the highest of the Japanese cypress stands. Hillslope 2 had relatively dense fern and evergreen shrub understory. Hillslope 3 was covered by deciduous forest for at least the last 40–50 years; understory consisted of relatively dense fern and shrub cover. Understory vegetation cover was <10% in hillslope 1, whereas in both hillslopes 2 and 3 it was nearly 100% (Table 1). The soil surface was covered by 2 to 3 cm of leaf and organic litter in hillslopes 2 and 3.

Figure 3.

Soil hydrophobicity, saturated hydraulic conductivity, bulk density, organic matter content, and soil particle size distribution on study hillslopes.

Table 1. Site Information of Various Study Plots
HillslopeSizeWidth (m)Length (m)Area (m2)Slope Gradient (deg)Stand Density (stems/ha)Forest TypeUnderstory Vegetation Cover (%)Understory Vegetation
1Large8.024.6118.043.04500Japanese cypress10Sparse
 Small0.51.70.8532.04500Japanese cypress10Sparse
2Large7.923.5120.041.01500Japanese cypressnearly 100Fern and evergreen
 Small0.51.60.8037.01500Japanese cypressnearly 100Fern and evergreen
3Large8.026.0130.041.0naDeciduousnearly 100Deciduous
 Small0.51.50.7343.0naDeciduousnearly 100Deciduous

[14] To examine soil physical properties of the hillslopes, three undisturbed soil core samples were collected in areas at depths of 5, 12.5, 20, and 40 cm. Saturated hydraulic conductivity (Ks), organic matter content, and particle size distribution were measured on the basis of these samples. Cores were taken using 100 cm3 steel cylinders with cross-sectional areas of 20 cm2 and heights of 5.1 cm. Soil cores were saturated in the laboratory for 48 h. Then Ks was measured using the falling head test [Reynolds et al., 2002]. Soil samples were oven dried at 105°C and weighed to measure bulk density, and then a subsample was ashed for 2 h at 550°C to estimate organic matter content. Oven-dried samples were then sieved into size classes of 8.0, 4.0, 2.0, 1.0, 0.5, 0.25, and 0.106 mm to determine particle size distribution.

[15] Separate bulk soil samples from the surface and depths of 5, 10, and 20 cm were collected for water repellency tests. All samples were dried at 60°C for 48 h in the laboratory before measuring the standardized soil moisture conditions [de Jonge et al., 1999]. Samples were equilibrated at ambient laboratory conditions for two days because soil water repellency is affected by air temperature and relative humidity. The sensitivity was examined using the critical surface tension (CST) test, in which drops of different concentration ethanol solutions were placed on the soil surface and the time required for the drops to infiltrate was measured [Watson and Letey, 1970]. Solutions with volumetric ethanol concentrations of 0, 1, 3, 5, 8.5, 16, and 30% were used to examine the range of hydrophilic conditions [Doerr et al., 1998]. Five drops of solution were placed on the soil surface using a micropipette. If all drops failed to infiltrate within 5 s, solutions with successively higher concentrations were applied. When all the drops infiltrated within 5 s, the ethanol concentration was taken as the resultant score.

[16] On each of three hillslopes, we installed large and small plots. Overland runoff was assumed to flow perpendicular to contour lines on the planar hillslope plots. Large plots represented entire hillslope segments, whereas small plots comprised a limited portion of the hillslope. Large plots were 8 m wide and 24 to 27 m long, located on planer hillslopes, and overland flow was assumed to occur perpendicular to contour lines. Slope length among plots varied depending on the distance from the ridge line to the valley bottom (Table 1 and Figure 3). The sides and upper extent of all plots were unbordered to avoid altering flow paths at the hillslope scale [Williams and Bonell, 1988]. Gradient of plots ranged from 41 to 43°; thus, the projected plot areas ranged from 108 to 130 m2. At the lower boundary of these plots, plastic troughs were installed parallel to the ridge lines and slope contours to collect surface runoff and sediment. Flexible aluminum flashing was inserted between the soil surface and the 1–3 cm depth and about 3 cm laterally in the soil matrix to facilitate the effective routing of runoff into the troughs. Because the flashing was only inserted into a relatively short slope distance (3 cm) of the soil, it did not affect infiltration capacities and upslope flow paths within the plots. The troughs were covered with small plastic roofs to avoid direct precipitation into the troughs. Water collected in the trough was directed through a drop box 45° V notch weir attached to the downslope end of the trough. Water stage near the inlet of the V notch weir was monitored every 5 min using capacitance water stage data loggers (TruTrack); stage was calibrated to discharge. Monitoring of large plots 1 and 2 initiated on 5 May 2004; monitoring of large plot 3 began in November 2004. Tensiometers were installed adjacent to the large plot 1 along the hillslope of catchment 5 in May 2005 to monitor pore pressure near the soil surface (5 cm depth). Tensiometer 1 was located 3 m from the valley bottom. Tensiometers 2 and 3 were installed 3 and 8 m apart, upslope of Tensiometer 1.

[17] Small plots were 0.5 m wide and 2 m long and located 20–30 m away from large plots on similar slope gradients and understory vegetation conditions (Figure 3). Trees grew around these small plots, but not within the plots (Table 1). Unlike large plots, small plots were bordered by plastic sheets both along the sides and upper boundary. All surface water was collected in gutters installed at the lower end of small plots and was directed into 200 L tanks. Similar to the large plots, flexible aluminum flashing was installed to capture overland flow. Water levels in tanks were measured every 5 min using capacitance water stage data loggers. Discharge at 5 min intervals was calculated on the basis of respective changes in water levels. Monitoring started in July 2004 in all small plots. Precipitation was measured by a tipping bucket rain gauge located in an open area 50 m from the outlet of the 4.9 ha study catchment. Total runoff volume (mm) was divided by the projected area of hillslope plots. Runoff coefficient was calculated as total runoff depth divided by total storm precipitation. Runoff coefficients may have be underestimated because we did not correct for interception losses of rainfall.

[18] For all storms, we estimated runoff volume from large plots on the basis of the runoff connectivity assumption (Figures 1 and 2) and runoff data from small plots. Estimated runoff from large plots was compared to observed values. The ratio between observed (Q) and estimated (Q*) overland flow from large plots expressed as the projected extended area of mean runoff length implies connectivity (C):

equation image

[19] A runoff connectivity ratio >1 indicates that overland flow on the large plot is more connected than on the small plot (runoff transfer dominant), whereas a connectivity ratio <1 implies that overland flow is less connected in the large plot than the small plot (infiltration dominant). A runoff connectivity ratio equal to 1 indicates that the extent of runoff connectivity is similar between small and large plots. Connectivity ratio was estimated on the basis of bulk overland flow volumes and values for each time step (i.e., every 5 min).

[20] Relations between runoff connectivity and both precipitation and antecedent soil moisture conditions were examined using correlation analysis. Total precipitation, maximum 5-, 20-, and 60-min storm intensity, and 7-day antecedent precipitation indices (API7) were used. API7 was selected as an indicator of moisture conditions near the soil surface. Threshold precipitation was estimated by total amount of precipitation required for the first response of overland flow in each event. Pearson's product moment correlation analysis was used for runoff connectivity and soil and antecedent precipitation indices. All variables in this analysis were log (x + 1) transformed to meet the assumption of normality and variable equality.

[21] Selection of small and large plots on hillslopes potentially affects scaling phenomenon of overland flow generation. For hillslope-scale studies, small plots tend to be selected to minimize the variability of hillslope conditions (i.e., smoother surfaces), whereas greater variability of soil surface conditions and microtopography was inherent in the large plots. The stem and canopy distribution around plots may affect variability of rainfall intensity and thus may alter scaling phenomenon; this is especially true for small plots. Indeed, it is impossible to select identical conditions (i.e., equal soil surface conditions and precipitation input regimes) between small and large plots for all hillslopes. Despite such difficulties in examining scaling effects of overland flow, we included all potential variability in our experimental design. We assume that if the variability due to plot selection and rainfall input is a major controlling factor of runoff generation, patterns of runoff from small and large plots may not be detected.

4. Results

4.1. Soil Properties

[22] Saturated hydraulic conductivity (Ks) ranged from 570 to 6920 mm h−1 (corresponding 0.016 to 0.19 cm s−1, Figure 3). No significant differences in Ks values were found among sample depths in hillslopes 1 and 2, whereas relatively high Ks values occurred near the soil surface in hillslope 3. Bulk density of soils ranged from 0.62 to 1.11 g cm−3, with relatively lower values near the surface due to the higher soil organic matter near the surface (Figure 3). Although hillslope 1 had sparse understory and litter cover, organic matter in the soil surface was similar to the other hillslopes. Surface soils consisted of finer materials than deeper soils for all hillslopes. Surface and near-surface soil had “very strong” hydrophobic conditions in hillslopes with Japanese cypress forest cover and “slightly” hydrophobic conditions in the deciduous forest. Soil depths ≥10 cm did not show any indication of hydrophobicity at all sites.

4.2. Rainfall and Runoff in Small and Large Plots

[23] Thirty five rainfall events occurred during the 2004–2005 study period; 80% of these events were in the typhoon season of 2004 (Figure 4). Annual precipitation in 2004 was 3201 mm, the highest on record (Kayumi climate station; ≈9 km northwest of the study site) since 1976. Indeed, many of typhoons struck Japan in 2004. In contrast, annual precipitation in 2005 (1383 mm) was the fifth lowest in the 20-year climate record. The Baiu rainy season from May to June 2005 was relatively dry compared to the long-term average (Figure 4). The 7-day antecedent precipitation index remained relatively low through winter and spring of 2005. Thus, our runoff data were collected for a range of wet and dry conditions. Overland flow and runoff coefficients varied seasonally. Runoff coefficients gradually decreased during the 2004 typhoon season especially in the large plots (Figure 4d). Seasonal patterns of runoff amount from small plots were less significant compared to large plots.

Figure 4.

(a) The 1-h precipitation and cumulative precipitation during the monitoring period from June 2004 to September 2005, (b) the 7-d antecedent precipitation index, (c) total runoff amount (mm) in small (open circles) and large (solid circles) plots of hillslope 1 and cumulative runoff amount, (d) runoff coefficients in large plot 1, (e) runoff coefficient in small plot 1 (Japanese cypress with sparse understory vegetation), and (f) changes in ratio of observed and estimated values of overland flow in hillslope 1 with Japanese cypress and sparse understory vegetation.

[24] Overland flow was observed in both small and large plots, regardless of forest conditions. Total overland runoff during storms was 2 to 10 times higher from small plots compared to large plots (Figure 5). The amount of overland flow, especially from small plots, tends to increase disproportionately with increases in total storm precipitation (Figure 5). For all hillslopes, small plots produced more overland flow per unit area than large plots, although variations of overland flow among small plots and among events were also greater than variations in large plots (Figure 5). For large plots, the greatest amount of overland flow occurred on the hillslope with Japanese cypress and sparse understory vegetation followed by Japanese cypress with denser understory vegetation and finally deciduous forest. For 10 storms in which overland flow was measured in all large plots, mean runoff coefficients for large plots 1, 2, and 3 were 10.4, 2.6, and 0.7%, respectively; however most of the storms that produced high runoff coefficients from the sparse understory hillslope (plot 1) were relatively small (<75 mm). During these same 10 storms, mean runoff coefficients for small plots 1, 2, and 3 were 26.5, 18.2, and 10.5%, respectively. The differences in runoff coefficients between large and small plots were the greater in deciduous hillslopes compared to Japanese cypress hillslopes.

Figure 5.

Total storm precipitation and total overland flow in small and large plots with different vegetation types. Dashed lines indicate 10% values for runoff coefficients.

4.3. Estimation of Runoff Connectivity

[25] Estimated runoff distances (lcs) from large plots 1, 2, and 3 for different storms were 0.02 to 0.95 m, 0.01 to 1.05 m, and 0.05 to 1.84 m, respectively. Differences between estimated (Q*) and observed (Q) overland flow in large plots implies extent of runoff connectivity (Figures 1 and 6) . Observed runoff from the Japanese cypress hillslope with sparse understory (hillslope 1) was greater than estimated runoff. This result implies that the mean length of overland flow transfer was underestimated when the values are compared to the large plot. In contrast, overland flow predicted from the hillslope with deciduous forest cover (hillslope 3) was slightly, but not significantly, higher than observed runoff (Figure 7). Thus, mean length of overland flow estimated in the large plot (on the basis of small plot runoff) was slightly overestimated compared to observed overland flow from the large plot. For hillslope 2 with Japanese cypress and fern understory vegetation, measured and predicted runoff volumes and variations were similar (Figure 7), implying that mean runoff length in the large plot was similar to observed values. Thus, our assumption based on Figure 2 was valid for this hillslope. On the basis of these estimates, runoff connectivity calculated by equation (7) was greatest in hillslope 1 followed by hillslopes 2 and 3.

Figure 6.

Responses of overland flow in small and large plots of hillslope 1 during a storm event on 6–7 September 2005 with rather wet antecedent moisture. Shown are (a) precipitation, (b) runoff of small and large plots, (c) estimated and observed overland flow based on the runoff discontinuity assumption, and (d and e) pressure head responses at the 5 and 10 cm depths, respectively, along hillslope 1.

Figure 7.

Observed and estimated overland flow volume (mm) in large plots. Estimated overland flow was derived on the basis of runoff discontinuity assumption (Figure 2). Error bars indicate 5th and 95th percentiles of samples.

[26] Connectivity ratio (estimated values divided by observed values of runoff in large plots) in hillslope 3 significantly increased with increasing in total runoff and rainfall intensity (Table 2). Rainfall intensities of 20 and 60 min duration were most significantly correlated to runoff connectivity in hillslope 3. Runoff connectivity significantly decreased with increasing API7 for hillslopes 2 and 3. No significant correlation was found between runoff connectivity and rainfall parameters for hillslope 1 (Table 2). Despite this lack of correlation, runoff connectivity ratio tended to be greater during the relatively dry period from spring to early summer of 2005 in hillslope 1 (Figure 4f).

Table 2. Summary of Correlation Analysis Between Runoff Connectivity and Rainfall and Soil Moisture Conditiona
 Hillslope 1Hillslope 2Hillslope 3
Correlation Coefficientp ValueCorrelation Coefficientp ValueCorrelation Coefficientp Value
  • a

    Analysis were conducted based on data from May and September 2005. Significant correlations are shown in bold.

Total precipitation−0.070.820.400.110.540.03
Maximum 5 min intensity0.050.850.170.520.610.01
Maximum 20 min intensity−0.140.610.230.360.65p < 0.01
Maximum 60 min intensity−0.160.560.310.240.72p < 0.01
API7−0.210.450.540.030.510.04

[27] Differences in estimated and observed runoff values (runoff connectivity ratio) also changed from the beginning to the end of storm events depending on rainfall intensity (Figures 6 and 8) . During a storm on 6–7 September 2005, predicted overland runoff tended to agree to the observed runoff when rainfall intensity was low to moderate (<3 mm per 5 min, Figure 6c). However, observed overland flow was 1.5 to 2.0 times greater than estimated values during intense precipitation (nearly 6 mm per 5 min.). During such intense precipitation with high concurrent overland flow, pressure head at the 5 cm depth near the bottom of hillslope 1 (tensiometer 1) was positive (Figure 6c). Predicted overland flow was consistently larger than observed values for the deciduous hillslope (Figure 8).

Figure 8.

Runoff from small and large plots of the deciduous forest during a storm on 6–7 September 2007 with rather wet antecedent moisture. The bottom plot shows an expanded view of observed and estimated runoff from the large plot in hillslope 3.

[28] Short, intense precipitation contributed to increases in connectivity of overland flow. Although antecedent soil moisture conditions prior to the 5–7 September 2005 storm were higher than the 28–29 September 2004 event, prolonged overland flow occurred during intense rainfall bursts during both events. Total and maximum 1-h precipitation during the 4 October 2004 storm was similar to the characteristics of the 28–29 September 2004 event (Figure 9). Short-term maximum precipitation intensity (based on 5 min intervals) in the October event was much smaller (maximum 3.5 mm per 5 min) than in the September event (maximum 7.5 mm per 5 min). Larger differences noted in the September event indicate that the greater connectivity of overland flow during the September event is likely attributable to higher short-term rain intensity.

Figure 9.

Responses of overland flow in small and large plots of Japanese cypress with sparse understory vegetation (hillslope 1) during storm events on 28–29 September and 4 October 2004. Both events had similar total and maximum 1-h precipitation and relatively dry antecedent soil moisture conditions. Short-term storm intensities differed: the event on 4 October 2004 never exceeded 4 mm in 5 min.

[29] Correlation coefficients for the relationship between connectivity ratio and 20–25 min moving average precipitation were highest for the September storm event (Figure 10a). This result suggests that the upslope extension of runoff connectivity was associated with progressive increases in 20 to 25 min precipitation. However, as observed in the October event, the upslope extension was small when rainfall intensity remained low throughout the storm. Mean 20-min precipitation in the range of 1 to 1.5 mm appeared to be the threshold for the upslope extension of the overland flow network (Figure 10b).

Figure 10.

Relationships between precipitation parameters and ratio between observed and estimated values. The legend applies to both plots.

5. Discussion

5.1. Factors Affecting Overland Flow Generation

[30] For all of hillslopes with the large and small plots, runoff responded simultaneously to precipitation. The duration of overland flow after the development of surface runoff was also relatively short (very rapid recession). No delayed runoff (i.e., kinematic wave assumption) related to slope length was detected in our 5 min monitoring intervals. Because the soil matrix of nearby hillslopes was not saturated on the basis of tensiometric observations, all of these runoff responses suggest the occurrence of infiltration-excess overland flow in these hillslopes.

[31] We measured overland flow irrespective of vegetation/ground cover condition and despite high measured saturated hydraulic conductivities of surface soils (Figure 3). Spatially distributed and representative values of saturated hydraulic conductivity at the hillslope scale were difficult to obtain because core sample size is too small to capture the spatial heterogeneity of hydraulic conductivity. Potential factors promoting overland flow are (1) water repellency; (2) flow within and across fallen leaves, branches, and horizontal root network; and (3) combinations of these factors. The highest soil water repellency was estimated near the soil surface layer between the litter and mineral horizon (Figure 3) [Miyata et al., 2007]. Hydrophobicity near the soil surface and resultant surface flow was also reported in a subalpine region of British Columbia [Barrett and Slaymaker, 1989], eastern Canada [Buttle and Turcotte, 1999], and a mountainous area of Switzerland [Scherrer et al., 2007]. Flow within and across fallen leaves and branches may also promote overland flow because we installed aluminum flashing 1 to 3 cm under soil surface. This near-surface phenomenon is called quasi-overland flow [Burch et al., 1989; Badoux et al., 2006] or biomat flow [Sidle et al., 2007]. Stemflow likely contributes only a trivial amount to overland flow because it typically comprises just 1 to 10% of precipitation [Levia and Frost, 2003].

[32] The high variability in runoff from small plots was potentially associated with seasonal variations in soil surface conditions. Because certain soil physical conditions (e.g., soil hydrophobicity, lack of surface roughness, surface soil sealing, ponding) and vertical infiltration tend to dominate within small plots, changes in such conditions also induce variations in runoff at this scale, possibly masking the differences in runoff among small plots with different vegetation cover. Spatial distribution of precipitation can strongly affect runoff from small plots [Wilson et al., 1979]. Miyata et al. [2007] monitored storm runoff from three 1 × 2 m plots in our Japanese cypress forest with sparse understory (near hillslope 1); one plot had twice the runoff compared to the other plots on the same hillslope. Additionally, formation of soil surface crusting and sealing may affect overland flow generation and infiltration especially on hillslopes with sparse understory vegetation [Onda and Yukawa, 1994; Nanko et al., 2008].

5.2. Scaling Effects and Runoff Connectivity of Overland Flow

[33] Differences in runoff coefficients between large and small plots showed scaling effects related to runoff generation from forested hillslopes. Scaling effects are associated with differences in the opportunities for overland flow transfer and infiltration. Mean runoff coefficients from our hillslopes were much smaller than agricultural and grass lands, likely due to the more heterogeneous soil surface conditions and greater opportunities for infiltration on forest hillslopes (Figure 11). Our small plots necessarily had relatively smoother surfaces than large plots to facilitate runoff collection at that scale [Sidle et al., 2007]. Long plots allow more time for overland flow to be absorbed by the soil than shorter plots because of greater hydraulic roughness and detention [Duley and Ackerman, 1934; Mutchler and Greer, 1980]. Small surface depressions also contribute to the spatial variability of ponding and infiltration [Dunne et al., 1991; Darboux et al., 2001; Vigiak et al., 2006]. Spatial variations of soil moisture, soil water repellency, and soil physical properties (e.g., clay composition) alter the timing of more continuous flow and surface pathways [Burch et al., 1989; Wood et al., 1986; Imeson et al., 1992; van de Giesen et al., 2000].

Figure 11.

Scaling effects of overland flow generation for different hillslope lengths. Values from Sharma [1986], Duley and Ackerman [1934], and Lal [1983] are mean annual runoff coefficients. Values from Joel et al. [2002] and our study were based on the mean of all storm runoff events. Values from Parsons et al. [2006] were estimated from mean runoff coefficients of Parsons et al. [2006, Figure 5].

[34] Although the amount of overland flow among small plots was similar, runoff from large plots differed among hillslopes. Runoff differences between large and small plots were more pronounced in the deciduous forest than in Japanese cypress with sparse understory vegetation (Figure 4). Thus, the extent of scaling phenomenon associated with overland flow among different types of hillslopes is potentially associated with runoff connectivity in hillslopes. Runoff connectivity was greatly extended during short and intense periods of precipitation, especially in the forested hillslope with sparse understory vegetation, which produced more overland flow. Water tables approached the soil surface for relatively short periods during intense precipitation (Figure 9). Continuous flow occurred during short and intense rainfall producing water tables near the soil surface (5 cm depth) in lower portions hillslope 1 compared to midslope to upslope locations (Figure 6). The lower hydraulic roughness on slopes with sparse understory vegetation cover caused more rapid transfer of runoff from upper to lower slopes. In the deciduous hillslope, more infiltration may have occurred in preferential flow pathways within pockets of decomposed organic matter and soil cracks when flow occurred in shallow subsurface layers (biomat) because observed overland flow was smaller than estimated overland flow (Figures 7 and 8). The rough surface litter layer (containing branches and exposed roots) would reduce runoff velocity, promote more water storage, and increase infiltration into mineral soil, thus augmenting subsurface flow [Abrahams et al., 1994; Sidle et al., 2007].

[35] Despite the fact that intense precipitation during storms produced more connected overland flow, connectivity ratio was not significantly correlated to total precipitation and 1 h rainfall intensity in hillslopes 1 and 2 (Table 2). Because intensity of rainfall can induce crusting and sealing at the soil surface, the mean value and spatial variability of infiltration capacity may change during the rainy season. It was also possible that changes in soil hydrophobicity through the rainy season may alter initial infiltration capacity. All of these factors, which promote greater variability of runoff, especially on hillslopes with sparse understory vegetation in Japanese cypress forests, contributed to additional variation in the relationships between runoff connectivity and rainfall intensity among storms. In contrast, because the mineral soil surface was protected by a litter layer on the deciduous hillslope, the connectivity ratio was significantly correlated with rainfall patterns (Table 2). Both hillslopes with litter cover (hillslopes 2 and 3) exhibited decreased runoff connectivity during wet conditions (higher API7), thus implying greater redirection of runoff to subsurface pathways.

[36] The consistent patterns of runoff observed between large and small plots suggested that findings of scaling phenomenon in our study hillslopes were robust. Selection of hillslope sites may contribute additional variability in runoff and affect runoff patterns between small and large plots. If variability (e.g., soil surface condition) between small and large plots were pronounced, patterns of scaling effects (e.g., decreasing runoff with increasing slope length) may be overwhelmed by such variability. Despite the variability that exists because of selection of hillslope plots, inferences of scaling effects for runoff generation were detected on the basis of our experimental design. The other possible artifact that affects the results of field measurements is the bordered and unbordered plot installations. Our findings showed that estimated runoff distance (lcs) never exceeded the small plot length (maximum runoff distance was 1.84 m in hillslope 3). However, these calculated slope lengths of runoff cannot be viewed exactly as the distance from the base of the plots, but rather an indicator of the degree of connectivity of overland flow. Values of runoff from small plots appeared to represent discontinuous runoff on hillslopes. It is also possible that lateral overland flow (as opposed to perpendicular to slope contours) may not be included in the small plots because they were bordered. However, we assume that the dominant runoff in both small and large plots was perpendicular to contour lines.

5.3. Dynamic Runoff Connectivity of Overland Flow

[37] Our analytical approach to runoff generation in small and large plots facilitated the examination of scaling effects and changes in connectivity of overland flow. Although our estimation of runoff distances did not represent the actual runoff transfer distances on hillslopes, comparing observed and estimated runoff on the basis of our analysis indicates changes in runoff connectivity with intensity of precipitation. Earlier studies on scaling effects typically used bulk volumes of overland stormflow and/or annual overland flow [e.g., Sharma, 1986; Lal, 1983; Joel et al., 2002; Parsons et al., 2006]. These approaches were useful to identify patterns of scaling effects when study sites were compared among treatments (e.g., vegetation cover, hillslope gradient). Changes in infiltration and runoff characteristics during a storm event as shown by Horton [1933] are critical to assess the timing and extent of scaling effects related to overland flow generation [Wainwright and Parsons, 2002]. High-resolution rainfall and runoff data are essential for detecting scaling effects and connectivity of overland flow.

[38] We show that progressive increases in precipitation intensity largely contribute to the upslope extension of the overland flow network, particularly in hillslopes with Japanese cypress stands and sparse understory vegetation. Perched water tables may develop at shallow soil depths only during the high and intense precipitation periods on the basis of near surface soil water tension measurements (Figure 6). Short, intense rainfall causes a rapid rise in the water table, fills topographic depressions, and creates surficial preferential flow networks [Moore and Larson, 1979; Dunne et al., 1991; Abrahams et al., 1989]. Particularly, for dry conditions, subsurface preferential flow pathways are discontinuous [Sidle et al., 2001], thus promoting localized surface flow. Thus, overland flow networks appeared to develop in response to a threshold of intense precipitation. Most of the connected overland flow is transferred through such flow networks on hillslopes rather than transported as sheet flow over slopes. Despite the occurrence of overland flow and the extended network of flow, we found no significant erosion features such as rills and gullies on any hillslopes (Figure 3). Thus, the amount and concentration of overland flow was not sufficient to create erosion features because of the disconnected nature of overland flow runoff even on the Japanese cypress slopes with sparse understory vegetation. Rain splash erosion and transport may dominate under such conditions [Nanko et al., 2008].

[39] Soil surface conditions associated with different vegetation types alter the development of connected networks of overland flow. Although overland flow occurred both from Japanese cypress and deciduous forest hillslopes, greater opportunities for infiltration create more discontinuous overland flow on deciduous forest slopes and Japanese cypress slopes with dense understory vegetation. Therefore, flow occurring near the organic horizon – mineral soil interface may infiltrate deeper in the soil and may not contribute to catchment storm runoff [Badoux et al., 2006; Sidle et al., 2007]. Litter on the forest floor also provides a site for temporary water storage and infiltration [Putuhena and Cordey, 1996]. Additional soil moisture promoted by the litter layer may prevent the development of hydrophobicity. However, on our cypress hillslope with sparse understory, overland flow that occurred during short, intense rainfall potentially reached stream channels and may contribute to storm runoff.

6. Summary and Conclusions

[40] Scaling effects and connectivity of overland flow were investigated by comparing storm runoff from small and large plots established in different forest stand conditions during 2004 and 2005. We demonstrated dynamic changes in spatial extent and timing of overland flow generation by comparing different sizes of plots and various vegetation cover types. The main findings of this study are summarized as follows: (1) overland flow was observed in both small and large plots, regardless of forest conditions; (2) total overland runoff during storms was 2 to 10 times higher from small plots compared to large plots; (3) overland runoff from small plots was variable, likely associated with variable precipitation, soil surface conditions, and lack of connected subsurface preferential flow paths; (4) on the basis of our runoff discontinuity assumption, hillslopes with deciduous stands had greater opportunities for infiltration, whereas hillslopes with dense Japanese cypress and sparse understory cover potentially generated more overland flow; (5) overland flow on slopes with sparse understory vegetation was connected and higher during short and intense storm events; (6) the development of an overland flow network depended on the opportunities for overland flow transfer affected by rainfall intensity and soil conditions on hillslopes; and (7) patterns and changes in overland flow generation on hillslopes are important to understand hillslope-stream runoff linkages and sediment and nutrient dynamics in forested headwater catchments.

[41] While it is obvious that subsurface flow is still the dominant pathway in these Japanese cypress forests (i.e., as measured and assumed in most studies of forest hillslopes), our findings suggest that connected overland flow can occur during storms, especially on hillslopes with sparse understory and very thin organic horizons. The significance of such rapid pathways for catchment-scale peak flow has yet to be established. Connected overland flow may promote suspended sediment and nutrient export from hillslopes to stream channels, thus altering sediment and nutrient dynamics of headwater catchments. Management practices such as partial cutting and commercial thinning support understory vegetation growth by allowing more light to penetrate the forest canopy; these practices may be helpful to reducing overland flow if practiced without compaction or soil disturbance. Our runoff comparisons in different sizes of plots provide valuable information for the development of improved numerical and process-based models for overland flow generation from forest hillslopes and in small catchments. These findings provide a useful basis for assessing the impacts of forest management on hydrological processes in steep forested catchments, such as dense Japanese cypress forests and other monoculture and second-growth forests. Understanding the hillslope runoff response with respect to scaling effects at appropriate temporal and spatial scales is essential to comprehend catchment runoff and hillslope-stream linkages of headwater systems.

Acknowledgments

[42] This study was supported by Japan Science and Technology Agency (JST), Core Research for Evolutional Science and Technology (CREST) project entitled “Field and modeling studies on the effect of forest devastation on flooding and environmental issues” and by a grant (16380102) from the Japan Society for the Promotion of Science (JSPS) to R. C. Sidle. We thank Masayasu Ueno, Tewodros A. Taddesse, Sohei Kobayashi, Aurelian C. Trandafir, Alan D. Ziegler, Fumitoshi Imaizumi, Mika Yamao, and Yotaro Nishi for their assistance with field work. We appreciate the insightful comments of two anonymous reviewers.

Ancillary

Advertisement